Middle Generation Console Upgrade Discussion [Scorpio, 4Pro]

[Silent_Buddha]

If Scorpio features 6TF and 320GB/s then in comparison to the RX 480 its theoretically 16% faster in the core and has 25% more bandwidth - however that bandwidth has to be shared with the CPU. Since theoreticals don't fully translate into performance increases it's doubtful that Scorpio would be more than about 15-20% faster than the RX 480.

The 980Ti averages 33% faster than the RX 480 in modern games so we should expect it to perform a little better than Scorpio but nothing Earth shattering.

https://www.techpowerup.com/reviews/AMD/RX_480/24.html

Well, besides the fact that GTX 980 ti has 96 ROPs compared to 32 for the Rx480. And benchmarks seem to indicate that Rx480 is ROP limited (compare Rx480 to R9 390/390x). This is especially visible as resolution and/or quality settings increase where the Rx480 doesn't scale as well.

As well, that's even assuming Scropio is based on Polaris, which I personally think is doubtful.

As well, assuming that there aren't any rendering efficiencies to be had on console versus PC.

Regards,
SB

 

[pjbliverpool]

haha...anyway...I think Scorpio will perform similar to gtx 980ti in games.

Well, besides the fact that GTX 980 ti has 96 ROPs compared to 32 for the Rx480. And benchmarks seem to indicate that Rx480 is ROP limited (compare Rx480 to R9 390/390x). This is especially visible as resolution and/or quality settings increase where the Rx480 doesn't scale as well.

As well, that's even assuming Scropio is based on Polaris, which I personally think is doubtful.

As well, assuming that there aren't any rendering efficiencies to be had on console versus PC.

Regards,
SB

I'd expect Scorpio to be Vega based but I wouldn't expect Vega to be any more efficient than Polaris per flop or per GB/s. Or are you saying that you think it might be if Scorpio has a proportionally higher number of ROPS? Say 64?

 

[BRiT]

Or it could "punch above its weight" due to console API efficiencies.

 

[betternatethanlever]

If Scorpio features 6TF and 320GB/s then in comparison to the RX 480 its theoretically 16% faster in the core and has 25% more bandwidth - however that bandwidth has to be shared with the CPU. Since theoreticals don't fully translate into performance increases it's doubtful that Scorpio would be more than about 15-20% faster than the RX 480.

The 980Ti averages 33% faster than the RX 480 in modern games so we should expect it to perform a little better than Scorpio but nothing Earth shattering.

https://www.techpowerup.com/reviews/AMD/RX_480/24.html

"33%" is a pretty significant difference in real world performance. I myself am expecting 390 levels of performance (assuming it's not Vega), but 980 Ti seems out of reach.

 

[Silent_Buddha]

I'd expect Scorpio to be Vega based but I wouldn't expect Vega to be any more efficient than Polaris per flop or per GB/s. Or are you saying that you think it might be if Scorpio has a proportionally higher number of ROPS? Say 64?

I'm saying there's no way to say as there are too many variables in play.

1. It's definitely going to have more than 32 ROPs as 4K is one of the targets.
2. The architecture is in question.
3. Console performance can be tailored specifically for one architecture versus multiple. This is especially important as on PC most software will generally be optimized for NVidia hardware.
4. Console allows you to get much closer to the metal. Most games on PC are on Dx11. And there are virtually no Dx12/Vulkan games really outside of Doom. And even the developers of Doom mentioned they couldn't fully leverage Vulkan as the PC engine was still based on OpenGL. On the Dx12 side, Ashes comes the closest, but even that isn't an engine fully developed around Dx12. This makes performance comparisons difficult.

I expect it to be faster than GTX 980 ti. Performance will likely be somewhere between the PC performance of GTX 1070 and GTX 1080, if I were to take a wild stab in the dark without knowing any details of the console outside of the extremely generic information that we currently have. It's possible it may even punch above that considering that all aspects of the hardware can be exploited on console while it cannot be on PC. IE - Dx11/12 features actually used in games will be limited to what is common between Nvidia and AMD hardware. That means that performance will be lower in those on PC than if you could optimize purely for one or the other.

However, a big question mark is what CPU is going to be used and at what speed it runs. That will determine whether the CPU will potentially limit the overall performance of games to a greater or lesser degree.

Regards,
SB

 

[sebbbi]

"33%" is a pretty significant difference in real world performance. I myself am expecting 390 levels of performance (assuming it's not Vega), but 980 Ti seems out of reach.

Doom Vulkan patch gave RX 480 around 30% increase. Most of this came from GCN specific wave shuffle intrinsics and async compute. Console games have used these features extensively for years. 980 Ti (Maxwell) doesn't gain anything from async compute, and most PC devs will not be writing code with GPU specific intrinsics for it.

In addition to this, console games will be specially profiled and optimized for a single closed platform. Shader bottlenecks will be solved and workloads balanced to fill the GPU execution units precisely. I would say that closed platform gives at least 20% extra performance in top of these already listed gains. This is especially true for AAA console exclusives designed around the console GPU.

I'd say Scorpio matches 980 Ti pretty well. Even if it uses Polaris.

 

[betternatethanlever]

Doom Vulkan patch gave RX 480 around 30% increase. Most of this came from GCN specific wave shuffle intrinsics and async compute. Console games have used these features extensively for years. 980 Ti (Maxwell) doesn't gain anything from async compute, and most PC devs will not be writing code with GPU specific intrinsics for it.

In addition to this, console games will be specially profiled and optimized for a single closed platform. Shader bottlenecks will be solved and workloads balanced to fill the GPU execution units precisely. I would say that closed platform gives at least 20% extra performance in top of these already listed gains. This is especially true for AAA console exclusives designed around the console GPU.

I'd say Scorpio matches 980 Ti pretty well. Even if it uses Polaris.

Fair enough. Are you expecting this kind of performance at launch?

 

[pjbliverpool]

I expect it to be faster than GTX 980 ti. Performance will likely be somewhere between the PC performance of GTX 1070 and GTX 1080, if I were to take a wild stab in the dark without knowing any details of the console outside of the extremely generic information that we currently have. It's possible it may even punch above that considering that all aspects of the hardware can be exploited on console while it cannot be on PC.

You expect it might be able to punch above a 1080? If we say it will gain 20% more performance from the extra functional units over an RX480 (which given 6TF is only a 16% increase over the RX480 is pretty generous), that means it would need to punch above it's weight by about 50% just to match a 1080 in real world performance which exceeds the 480 by about 77% according to TPU. I'd be dubious about that under DX11 never mind DX12 which should be much more prevalent by the time Scorpio hits the market. Punching above it's weight by 25% would be required to match a 1070 which might be realistic.

Putting that into perspective, the Radeon HD7970 averaged around 50% faster than the R7 265 which was an almost perfect match for the PS4 specs. Does the PS4 match the 7970's performance in any cross platform games? And even if it does, that's presumably under DX11, can it match a 7970 under DX12/Vulcan? And would be expect Scorpio to get the same level of optimisation that the PS4 has had?

https://www.techpowerup.com/reviews/Sapphire/R7_265_Dual-X/24.html

 

[MrFox]

Not sure what CPU they can fit on a 14/16nm SoC, considering it's sharing die area and TDP with a 6TF gpu. Comparisons with PC cpus might be a bit optimistic.

 

[iroboto]

Not sure what CPU they can fit on a 14/16nm SoC, considering it's sharing die area and TDP with a 6TF gpu. Comparisons with PC cpus might be a bit optimistic.

I agree with this, but I think the return response would follow along similar lines of sebbbis post about profiling for 1 GPU. The coders are also profiling for 1 CPU.

They know how many threads are available, they know the feature set that's available and can completely tailor their code to take advantage of all these little things that most generic games would avoid.

Like honestly, for what the XBO and PS4 are, they are quite impressive and will continue to improve. Until all engines have completely rebuilt pipelines away from DX11 pipelines I don't think we've seen the upper limits of these consoles yet. But we're certainly seeing the upper limits of DX11 style games running on these consoles.

Sent from my iPhone using Tapatalk

 

[Silent_Buddha]

You expect it might be able to punch above a 1080? If we say it will gain 20% more performance from the extra functional units over an RX480 (which given 6TF is only a 16% increase over the RX480 is pretty generous), that means it would need to punch above it's weight by about 50% just to match a 1080 in real world performance which exceeds the 480 by about 77% according to TPU. I'd be dubious about that under DX11 never mind DX12 which should be much more prevalent by the time Scorpio hits the market. Punching above it's weight by 25% would be required to match a 1070 which might be realistic.

Putting that into perspective, the Radeon HD7970 averaged around 50% faster than the R7 265 which was an almost perfect match for the PS4 specs. Does the PS4 match the 7970's performance in any cross platform games? And even if it does, that's presumably under DX11, can it match a 7970 under DX12/Vulcan? And would be expect Scorpio to get the same level of optimisation that the PS4 has had?

https://www.techpowerup.com/reviews/Sapphire/R7_265_Dual-X/24.html

As I said, and keep repeating.

We don't know what architecture Project Scorpio will be using. If it's Vega then comparisons to Polaris are virtually worthless until we know exactly what Vega entails. IIRC, it has a graphics IP level (assuming Greenland is part of Vega) that indicates it may be a larger departure from Polaris than Polaris was from past GCN chips. It could also be very similar. Either way we don't know.

It's likely Rx 480 would perform better with higher quality settings and/or higher resolutions if it had 64 instead of 32 ROPs. Again it's something we don't know because it doesn't exist. Considering the focus on 4K for Project Scorpio, I wouldn't even exclude the possibility of higher than 64 ROPs.

Those are all things we have absolutely zero information about at the moment. Sebbbi has already shown a situation where a more efficient rendering API can provide a 30% boost. And even that isn't as efficient as what could be done on consoles and it can't even target the GPU as much as you can on consoles. Meaning, that is it out of the question that the above 2 could lead to another 10-20% performance gain on top of a potential 30% gain?

As I stated. If I were to have to take a wild stab in the dark, I'd expect it somewhere between 1070 and 1080. That represents the greatest possibility in my mind. Yet you focus on me stating that it might be possible to exceed that instead of focusing on the range that I expect it to fall into. Yes, it's possible that it might exceed the 1080, but I don't expect it to exceed the 1080. Why? Because we don't know anything at all about what is going to be in Project Scorpio other than some numbers without a point of reference for those numbers to give them relevance

Regards,
SB

 

[Globalisateur]

As I said, and keep repeating.

We don't know what architecture Project Scorpio will be using. If it's Vega then comparisons to Polaris are virtually worthless until we know exactly what Vega entails. IIRC, it has a graphics IP level (assuming Greenland is part of Vega) that indicates it may be a larger departure from Polaris than Polaris was from past GCN chips. It could also be very similar. Either way we don't know.

It's likely Rx 480 would perform better with higher quality settings and/or higher resolutions if it had 64 instead of 32 ROPs. Again it's something we don't know because it doesn't exist. Considering the focus on 4K for Project Scorpio, I wouldn't even exclude the possibility of higher than 64 ROPs.

Those are all things we have absolutely zero information about at the moment. Sebbbi has already shown a situation where a more efficient rendering API can provide a 30% boost. And even that isn't as efficient as what could be done on consoles and it can't even target the GPU as much as you can on consoles. Meaning, that is it out of the question that the above 2 could lead to another 10-20% performance gain on top of a potential 30% gain?

As I stated. If I were to have to take a wild stab in the dark, I'd expect it somewhere between 1070 and 1080. That represents the greatest possibility in my mind. Yet you focus on me stating that it might be possible to exceed that instead of focusing on the range that I expect it to fall into. Yes, it's possible that it might exceed the 1080, but I don't expect it to exceed the 1080. Why? Because we don't know anything at all about what is going to be in Project Scorpio other than some numbers without a point of reference for those numbers to give them relevance

Regards,
SB

Exceed the 1080? With AMD 6tflops tech? You can't be serious. More like between 1060 and 1070 at best. There won't be an i7 in Scorpio.

 

[pjbliverpool]

We don't know what architecture Project Scorpio will be using. If it's Vega then comparisons to Polaris are virtually worthless until we know exactly what Vega entails. IIRC, it has a graphics IP level (assuming Greenland is part of Vega) that indicates it may be a larger departure from Polaris than Polaris was from past GCN chips. It could also be very similar. Either way we don't know.

I haven't heard anything that suggests Vega will offer a significant efficiency increase over Polaris but I'd be interested to take a look at your sources on that.

Sebbbi has already shown a situation where a more efficient rendering API can provide a 30% boost.

Indeed but the comparison was made between API's on the PC. If we assume the PC will be running DX12 or Vulcan rather than DX11 which was the baseline for the performance increase (we are talking about a year from now afterall) then at least some of that 30% is annulled on Pascal which can take some advantage from async compute unlike Maxwell. I'd assume there are other aspects of DX12 which Pascal can also benefit from.

And even that isn't as efficient as what could be done on consoles and it can't even target the GPU as much as you can on consoles. Meaning, that is it out of the question that the above 2 could lead to another 10-20% performance gain on top of a potential 30% gain?

But how much do we expect the Scorpio GPU to be specifically targeted? My understanding is that all Scorpio games must run on the XBO which will likely have a much larger install base. So yes, games will specifically target the XBO but Scorpio will potentially just be treated like a PC - i.e. more raw horsepower to push XBO optimised code.

[Yet you focus on me stating that it might be possible to exceed that instead of focusing on the range that I expect it to fall into. Yes, it's possible that it might exceed the 1080, but I don't expect it to exceed the 1080. Why? Because we don't know anything at all about what is going to be in Project Scorpio other than some numbers without a point of reference for those numbers to give them relevance

I'm honestly not seeing what's so difficult to predict about Scorpios performance if we assume that the rumoured specs of 6TF and 320GB/s are accurate. Do you really expect such a massive departure from Polaris's per unit efficiency? That wouldn't be consistent with any of AMD's previous architecture updates.

 

[mrcorbo]

Based on what we know for PS4 Pro and what we can assume for Scorpio based on what will be manufacturable, neither machine will have significant CPU power. One could conclude from this that neither Sony or MS consider CPU power to be an important contributor to the graphical capabilities of these consoles and the games they will be running. So why should we? The only place where I can see a weak CPU being a limitation is in VR games for Scorpio specifically where, since these games won't be on the original XBone, you could potentially design those games around having a relatively powerful CPU if one were available.

 

[mrcorbo]

But how much do we expect the Scorpio GPU to be specifically targeted? My understanding is that all Scorpio games must run on the XBO which will likely have a much larger install base. So yes, games will specifically target the XBO but Scorpio will potentially just be treated like a PC - i.e. more raw horsepower to push XBO optimised code.

What about developers taking the CPU code from the XBOne build, but then taking the DX12 code path from their PC build and addding Scorpio optimizations to that instead?

 

[Shifty Geezer]

One could conclude from this that neither Sony or MS consider CPU power to be an important contributor to the graphical capabilities of these consoles and the games they will be running.

Or that they hadn't any other options. They have to use an APU. They have to use whatever CPU AMD can provide there. Ergo they have to use Jaguar (Puma), and short of adding a load more cores can only bump the clock for that. Which Sony did. So we could be looking at a system where the CPU choice isn't any choice at all.

 

[dobwal]

Not sure what CPU they can fit on a 14/16nm SoC, considering it's sharing die area and TDP with a 6TF gpu. Comparisons with PC cpus might be a bit optimistic.

Wasn't isn't a year or two ago where AMD was talking about an Exascale apu sporting 32 Zen cores, a Greenland based igpu with 32 GB of HBM?

If AMD believes it can produce something like that, then producing an 8 zen core apu with a 6 Tflop igpu should be a cakewalk. LOL.

 

[mpg1]

Isn't the 1070 pretty close to the 980ti...like almost the same in terms of performance?

No way it comes close to 1080 though.

 

[Cyan]

Isn't the 1070 pretty close to the 980ti...like almost the same in terms of performance?

No way it comes close to 1080 though.

the GPU should be okay, question is....what about the CPU?

Shannon Loftis from Microsoft publishing studios has said that 1st party launch games are going to run natively at 4K on Scorpio. She said:

"Any games we're making that we're launching in the Scorpio time frame, we're making sure they can natively render at 4K."

So I guess that some equilibrium in the APU should be expected.

http://www.usatoday.com/story/tech/...22/?siteID=je6NUbpObpQ-5NYtnfHgUKcOfI6IK4xUnA

 

[mrcorbo]

Or that they hadn't any other options. They have to use an APU. They have to use whatever CPU AMD can provide there. Ergo they have to use Jaguar (Puma), and short of adding a load more cores can only bump the clock for that. Which Sony did. So we could be looking at a system where the CPU choice isn't any choice at all.

But if the system was going to bottleneck on the CPU capability with any regularity, why not just shave off some CUs from the design and make a smaller chip that wouldn't bottleneck on the CPU as often?

My expectation is that Sony have enough CPU power in PS4 Pro and MS are going to have enough CPU power in Scorpio that developers will be able to fully engage their GPUs the majority of the time.